20161219-Gengarajoo-Thesis.pdf (4.6 MB)
Download fileTrust Evaluation Based Dynamic Task Allocation Framework for Multi Mobile Robots System
thesis
posted on 2016-12-20, 22:12 authored by Rishwaraj GengarajooThe application of
multirobot system (MRS) is becoming increasingly popular as it allows multiple
robots within the same system to pool resources together to achieve common
goals or objectives. A vital part of the MRS is the teamwork and cooperation
through the sharing of information or resources among the robots to optimize
their efforts in accomplishing given mission objectives. A critical part of the
teamwork effort is the ability to trust each other when executing any task to
ensure efficient and successful cooperation. In this PhD research, the
credibility or the trustworthiness of a robot is the main focus of this study. More
specifically, this research presents the development of a trust evaluation
model and a trust based control system framework for dynamic task allocation in
MRS.
The developed trust evaluation model and the integrated trust based control system framework allows a robot to give a task to and/or receiving a task from trustworthy robots as well as dynamically form centralized teams of trustworthy robots only. The trust evaluation model is developed using the theory of temporal difference learning, incorporated with a novel, modified concept of Markov games and integrated with an original heuristic computational model to estimate trustworthiness.
Simulation experiments are conducted to study and analyze the performance of the developed model against some of the recently reported models in the literature. The simulation experiments indicate that the developed model performs better than the literature models. In terms of accuracy, the developed model scored the lowest RMSD compared the literature models, indicating higher accuracy. The developed model is also 17.8% more efficient in estimating trustworthiness than the best model from the literature. The integrated trust model based Control System also proved to be able to distinguish between trustworthy and untrustworthy robots effectively when accepting a task and forms preferred centralized coalitions only with trustworthy robots. The trust evaluation model and the control system is subsequently implemented in real world MRS experiments. The experiments proved that the robots with the embedded trust evaluation model and control system are able to identify trustworthy robots during task allocation and dynamically form teams of trustworthy robots in real time to complete the given task.
The developed trust evaluation model and the integrated trust based control system framework allows a robot to give a task to and/or receiving a task from trustworthy robots as well as dynamically form centralized teams of trustworthy robots only. The trust evaluation model is developed using the theory of temporal difference learning, incorporated with a novel, modified concept of Markov games and integrated with an original heuristic computational model to estimate trustworthiness.
Simulation experiments are conducted to study and analyze the performance of the developed model against some of the recently reported models in the literature. The simulation experiments indicate that the developed model performs better than the literature models. In terms of accuracy, the developed model scored the lowest RMSD compared the literature models, indicating higher accuracy. The developed model is also 17.8% more efficient in estimating trustworthiness than the best model from the literature. The integrated trust model based Control System also proved to be able to distinguish between trustworthy and untrustworthy robots effectively when accepting a task and forms preferred centralized coalitions only with trustworthy robots. The trust evaluation model and the control system is subsequently implemented in real world MRS experiments. The experiments proved that the robots with the embedded trust evaluation model and control system are able to identify trustworthy robots during task allocation and dynamically form teams of trustworthy robots in real time to complete the given task.
